Power Estimation for Two-Sample Tests Using Balanced Resampling

Uniform bootstrap resampling as described by Efron(1979) and others is an assumption-free method that can be used for some inferential problems including power estimation for two-sample tests. However, it is inefficient, requiring thousands of replications to achieve any reasonable accuracy- The purpose of this paper is to extend the one-sample balanced resampling method put forward by Davison, Hinkley and Schechtman (1986) to two-sample problems and apply it to power estimation. This extension is tried on simulated data as well as on an real data from the Iowa 65+ Rural Health Study. The power of two-sample bootstrap Wilcoxon test and t-test is estimated for different location shift alternatives and sample sizes. The simulation studies show that the efficiency and reliability of the balanced resampling method are much better than that of the uniform resampling method. Also, the asymptotic theory gives the same results.